TODO.txt

1. define (rigorously) an uptrend
  - an 'uptrend' starts on a day which a buy at open, with an initial stop of (low - (high-low)) from the day before, and traded following the rules in #5 results in a profit of > 0.2R
2. write a script to run through a csv of SPY chart data and extract all dates that an uptrend started, then poop out the AARF of the X days before which have been normalized on the uptrend day (1-0) 
3. define a normalization method for turning the days leading up to the uptrend start into a vector
  - uptrend start day maps to high = 1, low = 0
  - previous day maps to high = (high - initial low) / (initial high - initial low) | low = (low - initial low) / (initial high - initial low)
    - where 'initial high' is the high on the start day
4. implement a naive-bayes learning script to build models for all reasonable sets of days before uptrend starts
5. test all models on DIA, QQQ, with a closing system to determine validity
  - at the close of each day, move the stop up 0.09R, unless any other rule applies
  - if the low of the day is > 0.4R, move the stop up to the low minus $0.02.
  - if the stop would advance beyond the day's low, merely advance it to the low minus $0.02.

parameters to tweak:
- # of days before an 'uptrend' to record (hunch = 5+/- 2) test 1 - 30
- how much to move the stop up each day intially (hunch = 0.09R) test 0.01R - 0.3R
- learn on just closing vs high, low, close

naive bayes
logistic regression
WEKA
dynamic linear models in R
markov switching.